Security audit
Security checks for vulnerabilities and agentic risk
The skill is low-risk from a security standpoint, but its documentation is inconsistent about local storage and the package appears incomplete.
Install only if you are comfortable with a skill that may store user-added passages in a local codex file if that optional feature is implemented. Treat the included documentation as incomplete because the Python engine and demo files it references were not present in this artifact.
The trigger phrase 'what is the address of' is overly broad and not clearly constrained to the Borges/Library-of-Babel domain. In an agent environment, this can cause accidental invocation on unrelated requests involving addresses, leading to wrong-tool activation, confusion, and possible unintended processing or persistence of user content through the codex-related features.
At line 10 the skill asserts 'No storage' and elsewhere frames codex access as a static read, yet the same document advertises adding entries that are 'persisted immediately.' This creates a deceptive data-handling representation that can undermine informed consent and safe use even if no direct code execution or exfiltration is shown.
The documentation makes strong privacy and persistence claims such as 'No database. No storage.' but later describes persisting user-provided entries to codex.json immediately. This mismatch can mislead users and downstream agents about data retention behavior, causing them to share text under false assumptions that nothing will be stored.
No suspicious patterns detected.
